__thread_pointer on powerpc

__thread_pointer on powerpc

[Mathieu Desnoyers]

Hi Florian,

In the recent patchset introducing rseq support in glibc, __thread_pointer
is wired up for powerpc.

Based on my interpretation of

https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/Local-Register-Variables.html

it appears that using "register void *__result asm ("r2")" is not sufficient
to guarantee that the compiler will indeed use this register. It's documented
as being just a hint that the compiler is free to optimize away.

I suspect we may want to add something like:

asm ("" : "=r" (__result));

just after the variable definition, to force the compiler to use this variable
as output to an inline asm.

Thoughts ?

Thanks,

Mathieu

-- 
Mathieu Desnoyers
EfficiOS Inc.
http://www.efficios.com

Re: __thread_pointer on powerpc

[Florian Weimer]

* Mathieu Desnoyers:

> Hi Florian,
>
> In the recent patchset introducing rseq support in glibc, __thread_pointer
> is wired up for powerpc.
>
> Based on my interpretation of
>
> https://gcc.gnu.org/onlinedocs/gcc-6.1.0/gcc/Local-Register-Variables.html
>
> it appears that using "register void *__result asm ("r2")" is not sufficient
> to guarantee that the compiler will indeed use this register. It's documented
> as being just a hint that the compiler is free to optimize away.
>
> I suspect we may want to add something like:
>
> asm ("" : "=r" (__result));
>
> just after the variable definition, to force the compiler to use this variable
> as output to an inline asm.

Good point.  I think we should use a global register variable instead.
I will try to fix this.

Thanks,
Florian